EE Seminar: Improved LP decoding of HDPC codes with Penalized ADMM
Speaker: Ilia Shulman
M.Sc. student under the supervision of Prof. Yair Be’ery
Wednesday, September 21st, 2016 at 15:30
Room 011, Kitot Bldg., Faculty of Engineering
Improved LP decoding of HDPC codes with Penalized ADMM
Abstract
In 2003 Jon Feldman showed that a relaxed version of the maximum likelihood decoding problem can be stated as a Linear Program (LP). Since then the research of this approach progressed significantly such that recent works even discuss efficient hardware implementation of LP decoders. In our research we focus on LP decoding of classic block codes, currently referred as High Density Parity Check (HDPC) codes in contrast to LDPC codes. In our talk we will present and analyze an improved LP decoder based on a Penalized Alternating Directions Method of Multipliers (P-ADMM) algorithm, and demonstrate it on various BCH codes. We show that the performance improvement achived with our proposed decoder is significantly better for HDPC codes than it is for LDPC codes. We investigate the time complexity of the P-ADMM and present the effect of a penalty coefficient parameter on the performance/complexity trade-off. Finally, we present a detailed time complexity comparison between our LP P-ADMM decoder and a state of the art Belief Propagation (BP) decoder. We conclude with some ideas for future research.
